Kimi K3 tops GPT-5.6 Sol in DocBench Arena

Kimi K3 completed every task in DocBench Arena, a document and presentation creation benchmark, with 400+ Redditors casting 3,000+ blind votes on 236 generated files. The post claims the model exposed GPT-5.6 Sol's weak spot.